High Dynamic Range Visual Content Compression

This thesis addresses the research questions of High Dynamic Range (HDR) visual contents compression. The HDR representations are intended to represent the actual physical value of the light rather than exposed value. The current HDR compression schemes are the extension of legacy Low Dynamic Range (LDR) compressions, by using Tone-Mapping Operators (TMO) to reduce the dynamic range of the HDR contents. However, introducing TMO increases the overall computational complexity, and it causes the temporal artifacts. Furthermore, these compression schemes fail to compress non-salient region differently than the salient region, when Human Visual System (HVS) perceives them differently. The main contribution of this thesis is to propose a novel Mapping-free visual saliency-guided HDR content compression scheme. Firstly, the relationship of Discrete Wavelet Transform (DWT) lifting steps and TMO are explored. A novel approach to compress HDR image by Joint Photographic Experts Group (JPEG) 2000 codec while backward compatible to LDR is proposed. This approach exploits the reversibility of tone mapping and scalability of DWT. Secondly, the importance of the TMO in the HDR compression is evaluated in this thesis. A mapping-free post HDR image compression based on JPEG and JPEG2000 standard codecs for current HDR image formats is proposed. This approach exploits the structure of HDR formats. It has an equivalent compression performance and the lowest computational complexity compared to the existing HDR lossy compressions (50% lower than the state-of-the-art). Finally, the shortcomings of the current HDR visual saliency models, and HDR visual saliency-guided compression are explored in this thesis. A spatial saliency model for HDR visual content outperform others by 10% for spatial visual prediction task with 70% lower computational complexity is proposed. Furthermore, the experiment suggested more than 90% temporal saliency is predicted by the proposed spatial model. Moreover, the proposed saliency model can be used to guide the HDR compression by applying different quantization factor according to the intensity of predicted saliency map